def fa(m0, n0, zs_norm, thetas, funcnum=2):
"""Calculates the matrix with the base functions for `w_0`
The calculated matrix is directly used to calculate the `w_0` displacement
field, when the corresponding coefficients `c_0` are known, through::
a = fa(m0, n0, zs_norm, thetas, funcnum)
w0 = a.dot(c0)
Parameters
----------
m0 : int
The number of terms along the meridian.
n0 : int
The number of terms along the circumference.
zs_norm : np.ndarray
The normalized `z` coordinates (from 0. to 1.) used to compute
the base functions.
thetas : np.ndarray
The angles in radians representing the circumferential positions.
funcnum : int, optional
The function used for the approximation (see function :func:`.calc_c0`)
"""
try:
import _fit_data
return _fit_data.fa(m0, n0, zs_norm, thetas, funcnum)
except:
warn('_fit_data.pyx could not be imported, executing in Python/NumPy'
+ '\n\t\tThis mode is slower and needs more memory than the'
+ '\n\t\tPython/NumPy/Cython mode',
level=1)
zs = zs_norm.ravel()
ts = thetas.ravel()
n = zs.shape[0]
zsmin = zs.min()
zsmax = zs.max()
if zsmin < 0 or zsmax > 1:
log('zs.min()={0}'.format(zsmin))
log('zs.max()={0}'.format(zsmax))
raise ValueError('The zs array must be normalized!')
if funcnum==1:
a = np.array([[sin(i*pi*zs)*sin(j*ts), sin(i*pi*zs)*cos(j*ts)]
for j in range(n0) for i in range(1, m0+1)])
a = a.swapaxes(0,2).swapaxes(1,2).reshape(n,-1)
elif funcnum==2:
a = np.array([[cos(i*pi*zs)*sin(j*ts), cos(i*pi*zs)*cos(j*ts)]
for j in range(n0) for i in range(m0)])
a = a.swapaxes(0,2).swapaxes(1,2).reshape(n,-1)
elif funcnum==3:
a = np.array([[sin(i*pi*zs)*sin(j*ts), sin(i*pi*zs)*cos(j*ts),
cos(i*pi*zs)*sin(j*ts), cos(i*pi*zs)*cos(j*ts)]
for j in range(n0) for i in range(m0)])
a = a.swapaxes(0,2).swapaxes(1,2).reshape(n,-1)
return a
def fw0(m0, n0, c0, xs_norm, ts, funcnum=2):
r"""Calculates the imperfection field `w_0` for a given input
Parameters
----------
m0 : int
The number of terms along the meridian.
n0 : int
The number of terms along the circumference.
c0 : np.ndarray
The coefficients of the imperfection pattern.
xs_norm : np.ndarray
The meridian coordinate (`x`) normalized to be between ``0.`` and
``1.``.
ts : np.ndarray
The angles in radians representing the circumferential coordinate
(`\theta`).
funcnum : int, optional
The function used for the approximation (see function :func:`.calc_c0`)
Returns
-------
w0s : np.ndarray
An array with the same shape of ``xs_norm`` containing the calculated
imperfections.
Notes
-----
The inputs ``xs_norm`` and ``ts`` must be of the same size.
The inputs must satisfy ``c0.shape[0] == size*m0*n0``, where:
- ``size=2`` if ``funcnum==1 or funcnum==2``
- ``size=4`` if ``funcnum==3``
"""
if xs_norm.shape != ts.shape:
raise ValueError('xs_norm and ts must have the same shape')
if funcnum==1:
size = 2
elif funcnum==2:
size = 2
elif funcnum==3:
size = 4
if c0.shape[0] != size*m0*n0:
raise ValueError('Invalid c0 for the given m0 and n0!')
try:
import _fit_data
w0s = _fit_data.fw0(m0, n0, c0, xs_norm.ravel(), ts.ravel(), funcnum)
except:
a = fa(m0, n0, xs_norm.ravel(), ts.ravel(), funcnum)
w0s = a.dot(c0)
return w0s.reshape(xs_norm.shape)
def calc_c0(path, m0=50, n0=50, funcnum=2, fem_meridian_bot2top=True,
rotatedeg=None, filter_m0=None, filter_n0=None, sample_size=None,
maxmem=8):
r"""Find the coefficients that best fit the `w_0` imperfection
The measured data will be fit using one of the following functions,
selected using the ``funcnum`` parameter:
1) Half-Sine Function
.. math::
w_0 = \sum_{i=1}^{m_0}{ \sum_{j=0}^{n_0}{
{c_0}_{ij}^a sin{b_z} sin{b_\theta}
+{c_0}_{ij}^b sin{b_z} cos{b_\theta} }}
2) Half-Cosine Function (default)
.. math::
w_0 = \sum_{i=0}^{m_0}{ \sum_{j=0}^{n_0}{
{c_0}_{ij}^a cos{b_z} sin{b_\theta}
+{c_0}_{ij}^b cos{b_z} cos{b_\theta} }}
3) Complete Fourier Series
.. math::
w_0 = \sum_{i=0}^{m_0}{ \sum_{j=0}^{n_0}{
{c_0}_{ij}^a sin{b_z} sin{b_\theta}
+{c_0}_{ij}^b sin{b_z} cos{b_\theta}
+{c_0}_{ij}^c cos{b_z} sin{b_\theta}
+{c_0}_{ij}^d cos{b_z} cos{b_\theta} }}
where:
.. math::
b_z = i \pi \frac z H_{points}
b_\theta = j \theta
where `H_{points}` represents the difference between the maximum and
the minimum `z` values in the imperfection file.
The approximation can be written in matrix form as:
.. math::
w_0 = [g] \{c_0\}
where `[g]` carries the base functions and `{c_0}` the respective
amplitudes. The solution consists on finding the best `{c_0}` that
minimizes the least-square error between the measured imperfection pattern
and the `w_0` function.
Parameters
----------
path : str or np.ndarray
The path of the file containing the data. Can be a full path using
``r"C:\Temp\inputfile.txt"``, for example.
The input file must have 3 columns "`\theta` `z` `imp`" expressed
in Cartesian coordinates.
This input can also be a ``np.ndarray`` object, with
`\theta`, `z`, `imp` in each corresponding column.
m0 : int
Number of terms along the meridian (`z`).
n0 : int
Number of terms along the circumference (`\theta`).
funcnum : int, optional
As explained above, selects the base functions used for
the approximation.
fem_meridian_bot2top : bool, optional
A boolean indicating if the finite element has the `x` axis starting
at the bottom or at the top.
rotatedeg : float or None, optional
Rotation angle in degrees telling how much the imperfection pattern
should be rotated about the `X_3` (or `Z`) axis.
filter_m0 : list, optional
The values of ``m0`` that should be filtered (see :func:`.filter_c0`).
filter_n0 : list, optional
The values of ``n0`` that should be filtered (see :func:`.filter_c0`).
sample_size : int or None, optional
An in specifying how many points of the imperfection file should
be used. If ``None`` is used all points file will be used in the
computations.
maxmem : int, optional
Maximum RAM memory in GB allowed to compute the base functions.
The ``scipy.interpolate.lstsq`` will go beyond this limit.
Returns
-------
out : np.ndarray
A 1-D array with the best-fit coefficients.
Notes
-----
If a similar imperfection pattern is expected along the meridian and along
the circumference, the analyst can use an optimized relation between
``m0`` and ``n0`` in order to achieve a higher accuracy for a given
computational cost, as proposed by Castro et al. (2014):
.. math::
n_0 = m_0 \frac{\pi(R_{bot}+R_{top})}{2H}
"""
from scipy.linalg import lstsq
if isinstance(path, np.ndarray):
input_pts = path
path = 'unmamed.txt'
else:
input_pts = np.loadtxt(path)
if input_pts.shape[1] != 3:
raise ValueError('Input does not have the format: "theta, z, imp"')
if (input_pts[:,0].min() < -2*np.pi or input_pts[:,0].max() > 2*np.pi):
raise ValueError(
'In the input: "theta, z, imp"; "theta" must be in radians!')
log('Finding c0 coefficients for {0}'.format(str(os.path.basename(path))))
log('using funcnum {0}'.format(funcnum), level=1)
if sample_size:
num = input_pts.shape[0]
if sample_size < num:
input_pts = input_pts[sample(range(num), int(sample_size))]
if funcnum==1:
size = 2
elif funcnum==2:
size = 2
elif funcnum==3:
size = 4
else:
raise ValueError('Valid values for "funcnum" are 1, 2 or 3')
# the least-squares algorithm uses approximately the double the memory
# used by the coefficients matrix. This is non-linear though.
memfac = 2.2
maxnum = int(maxmem*1024*1024*1024*8/(64.*size*m0*n0)/memfac)
num = input_pts.shape[0]
if num >= maxnum:
input_pts = input_pts[sample(range(num), int(maxnum))]
warn('Using {0} measured points due to the "maxmem" specified'.
format(maxnum), level=1)
ts = input_pts[:, 0].copy()
if rotatedeg is not None:
ts += deg2rad(rotatedeg)
zs = input_pts[:, 1]
w0pts = input_pts[:, 2]
#NOTE using `H_measured` did not allow a good fitting result
#zs /= H_measured
zs = (zs - zs.min())/(zs.max() - zs.min())
if not fem_meridian_bot2top:
#TODO
zs *= -1
zs += 1
a = fa(m0, n0, zs, ts, funcnum)
log('Base functions calculated', level=1)
c0, residues, rank, s = lstsq(a, w0pts)
log('Finished scipy.linalg.lstsq', level=1)
if filter_m0 is not None or filter_n0 is not None:
c0 = filter_c0(m0, n0, c0, filter_m0, filter_n0, funcnum=funcnum)
return c0, residues
[-cos(b)*sin(g),
(cos(a)*cos(g) - sin(a)*sin(b)*sin(g)),
(sin(a)*cos(g) + cos(a)*sin(b)*sin(g)), y0],
[sin(b), -sin(a)*cos(b), cos(a)*cos(b), z0]])
if __name__=='__main__':
import matplotlib.pyplot as plt
from _fit_data import fa
path = r'C:\clones\desicos\desicos\conecylDB\files\dlr\degenhardt_2010_z25\degenhardt_2010_z25_msi_theta_z_imp.txt'
m0 = 20
n0 = 20
c0, residues = calc_c0(path, m0=m0, n0=n0, sample_size=0.75)
theta = np.linspace(-pi, pi, 1000)
z = np.linspace(0, 1., 400)
theta, z = np.meshgrid(theta, z, copy=False)
a = fa(m0, n0, z.ravel(), theta.ravel(), funcnum=1)
w = a.dot(c0).reshape(1000, 400)
levels = np.linspace(w.min(), w.max(), 400)
plt.contourf(theta, z, w.reshape(theta.shape), levels=levels)
plt.gcf().savefig('plot.png', transparent=True,
bbox_inches='tight', pad_inches=0.05)